摘要
针对迭代动态规划计算量大,耗费时间长的特点,基于实验室搭建的PC机群,以消息传递库MPICH为编程工具,搭建一个并行计算平台,给出一种主从式的并行迭代动态规划算法,利用该算法对聚合物驱最优控制问题进行求解,并与串行计算结果进行对比。结果表明:在大规模的优化问题中并行迭代动态规划算法与串行算法结果一致,但表现出较高的并行效率和加速比;并行算法求解的效率受到主节点分配任务时是否均衡的影响。
Since the iterative dynamic programming （IDF） algorithm is complex and time-consuming, a parallel computing platform was built by using PC in the lab and using the message passing library MPICH as a programming tool. A masterslave algorithm of the parallelization IDP was applied to solve the optimal control problem of polymer flooding. The results obtained by the parallelization IDP were compared with those by the sequential algorithm. The results show that for a large scale optimization problem, the optimal solutions by the parallelization IDP agree well with those by the sequential algorithm. And the parallelization IDP is characterized by high efficiency and fast speed-up ratio. The computing efficiency of parallelization IDP is influenced by host node balance in task allocation.
出处
《中国石油大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2009年第3期167-171,174,共6页
Journal of China University of Petroleum(Edition of Natural Science)
基金
国家科技重大专项“高温高盐油田提高采收率技术”课题(2008ZX05011)
国家“973”项目(2004CB31800)
关键词
迭代动态规划
消息传递接口
最优控制
并行化
聚合物驱
iterative dynamic programming
message passing interface （ MPI ）
optimal control
parallelization
polymer flooding
